Generic, network schema agnostic sparse tensor factorization for single-pass clustering of heterogeneous information...

Area of Science:

• Data Mining
• Network Analysis
• Machine Learning

Background:

• Heterogeneous information networks are common and understanding their structure via clustering is crucial.
• Existing clustering methods often assume simple network schemas and can only process one object type at a time.

Purpose of the Study:

• To propose a novel, schema-agnostic clustering framework for heterogeneous information networks.
• To enable simultaneous clustering of multiple object types in a single pass.

Main Methods:

• Modeling heterogeneous information networks as a sparse tensor.
• Formulating clustering as an optimization problem solvable with Tucker decomposition.
• Employing an Alternating Least Squares (ALS) algorithm for efficient computation.

Main Results:

• The proposed STFClus framework effectively models heterogeneous information networks.
• STFClus outperforms existing state-of-the-art clustering algorithms on synthetic and real-world datasets.
• Demonstrated simultaneous partitioning of diverse object types into clusters.

Conclusions:

• STFClus offers a generally applicable, single-pass clustering solution for heterogeneous networks.
• The framework's schema-agnostic nature enhances its versatility.
• Sparse tensor factorization provides an efficient approach to complex network clustering.